CTCF Expression is Essential for Somatic Cell Viability and Protection Against Cancer

CCCTC-binding factor (CTCF) is a conserved transcription factor that performs diverse roles in transcriptional regulation and chromatin architecture. Cancer genome sequencing reveals diverse acquired mutations in CTCF, which we have shown functions as a tumour suppressor gene. While CTCF is essential for embryonic development, little is known of its absolute requirement in somatic cells and the consequences of CTCF haploinsufficiency. We examined the consequences of CTCF depletion in immortalised human and mouse cells using shRNA knockdown and CRISPR/Cas9 genome editing as well as examined the growth and development of heterozygous Ctcf (Ctcf+/−) mice. We also analysed the impact of CTCF haploinsufficiency by examining gene expression changes in CTCF-altered endometrial carcinoma. Knockdown and CRISPR/Cas9-mediated editing of CTCF reduced the cellular growth and colony-forming ability of K562 cells. CTCF knockdown also induced cell cycle arrest and a pro-survival response to apoptotic insult. However, in p53 shRNA-immortalised Ctcf+/− MEFs we observed the opposite: increased cellular proliferation, colony formation, cell cycle progression, and decreased survival after apoptotic insult compared to wild-type MEFs. CRISPR/Cas9-mediated targeting in Ctcf+/− MEFs revealed a predominance of in-frame microdeletions in Ctcf in surviving clones, however protein expression could not be ablated. Examination of CTCF mutations in endometrial cancers showed locus-specific alterations in gene expression due to CTCF haploinsufficiency, in concert with downregulation of tumour suppressor genes and upregulation of estrogen-responsive genes. Depletion of CTCF expression imparts a dramatic negative effect on normal cell function. However, CTCF haploinsufficiency can have growth-promoting effects consistent with known cancer hallmarks in the presence of additional genetic hits. Our results confirm the absolute requirement for CTCF expression in somatic cells and provide definitive evidence of CTCF’s role as a haploinsufficient tumour suppressor gene. CTCF genetic alterations in endometrial cancer indicate that gene dysregulation is a likely consequence of CTCF loss, contributing to, but not solely driving cancer growth.

Our previous studies first demonstrated the growth inhibitory effects of CTCF in vitro [24] and subsequently confirmed that CTCF acts as a tumour suppressor gene in vivo by suppressing tumour growth [25]. Isolated CTCF mutations have been identified in breast, prostate and Wilms' tumours [26] and acute lymphoblastic leukaemia [27]. However recent cancer genome studies have revealed the extensive somatic mutations occurring in CTCF [28]. CTCF has been classified as a significantly mutated gene owing to its high frequency of mutation and deletion in endometrial cancer [29]. CTCF mutations are detected in 35% of endometrial carcinomas exhibiting microsatellite instability (MSI), and in 20% of MSI-negative tumours [30]. One report describing 17 oncogenic signatures in cancer, defines one signature, M5, as comprising MSI-positive endometrioid cancers and some luminal A breast cancers. In this subset of endometrioid and breast cancers, CTCF mutations were identified in 40% of samples including inactivation of specific zinc fingers (ZFs) of CTCF that would lead to altered DNA binding [31]. We since revealed that CTCF genetic alterations have a pro-tumourigenic effect in endometrial cancer by altering cellular polarity and enhancing cell survival [32].
Genetic lesions in CTCF, whether heterozygous deletion, nonsense, frameshift or even missense zinc finger (ZF) mutations, can all result in CTCF haploinsufficiency. In endometrial cancer, CTCF mRNA transcripts expressed from alleles containing nonsense or frameshift mutations are subjected to nonsense-mediated decay [30], [32]. Somatic missense mutations in residues critical for CTCF ZF binding to DNA can result in selective loss of binding to some CTCF target sites, but not all [26], indicating the functional implications of incomplete loss of CTCF binding in cancer is unclear.
Loss of heterozygosity (LOH) at 16q22 can lead to CTCF haploinsufficiency and IGF2 upregulation in Wilms' tumours [33]. To date, modelling the full impact of CTCF haploinsufficiency on CTCF's tumour suppressor function has not been previously examined.

Page 6 of 37
In this study we assessed several genetic models of CTCF haploinsufficiency to reveal in detail the impact of heterozygous loss of CTCF in somatic cells, whole mice and human endometrial cancer.
Depletion of CTCF expression in K562 erythroleukaemia cells using shRNA knockdown or CRISPR/Cas9-mediated targeting of CTCF decreased cellular proliferation. In vivo, Ctcf heterozygosity negatively impacted the growth and gross development of mice. However, p53 shRNA-immortalised Ctcf +/mouse embryonic fibroblasts (MEFs) were functionally distinct from wild type (WT) MEFs by exhibiting increased cellular growth and other known cancer hallmarks.
Importantly, we were unable to generate Ctcf nullizygous MEFs after CRISPR/Cas9 genome editing confirming that CTCF is absolutely essential for somatic cell viability. Finally, we examined curated human endometrial carcinoma genomic data and observed that CTCF haploinsufficiency contributed to the transcriptional dysregulation of specific loci as well as inducing a unique gene signature in human cancers.

Cell lines
Human erythroid leukaemia (K562) cells were grown in RPM1 1640 medium while human embryonic kidney (HEK293T) and mouse embryonic fibroblast cells were cultured in DMEM.

Retroviral and lentiviral transduction
Viral supernatants were produced by calcium phosphate transfection of HEK293T cells: with pJK3, pCMVTat and pL-VSV-G packaging plasmids used to produce retroviruses; and pRSV-Rev, pMDLg/p.rre and pMD2.VSV-G used to package lentiviruses. Viral supernatants collected after 24-48 h were 0.45 μM-filtered and snap-frozen or concentrated by ultracentrifugation for 2 h at 20,000 rpm in a SW28 Beckman rotor. Viral supernatant was resuspended on ice in 10% (v/v) FCS/DMEM at 1/100th of the original volume. Adherent cells (1-5×10 5 /well) were seeded in 6-well plates before addition of fresh medium containing viral supernatant (~5×10 5 transducing units) and Polybrene (4 μg/mL; Sigma) and 'spin-oculated' for 90 min at 1,500 rpm. The supernatant was replaced with medium 12 h post-transduction and fluorescent cells were purified 24 h later by fluorescence activated cell sorting (FACS; >95% purity on re-analysis) using a FACS Influx (Becton Dickinson, BD). K562 cells (~5x10 5 /mL) in 1 mL medium with 4 μg/mL Polybrene were placed in a 5 mL capped FACS tube and transduced with viral supernatant for 90 min by 'spin-oculation'. The cells were resuspended and incubated at 37 °C for 4 h before removal of viral supernatant. For in vitro assays, cells were either plated out immediately or allowed to recover after sorting for 48-72 h in medium containing 100 μg/mL Normocin (InvivoGen).

CRISPR/Cas9 genome editing, validation and molecular genetic analysis
SgRNAs targeting the first coding exon of human CTCF and mouse Ctcf (exon 3) were designed using the Zhang lab CRISPR design tool (crispr.mit.edu). SgRNAs targeting the adeno-associated virus integration site 1 (AAVS1) and the Rosa26 locus were used as negative control guides in human and mouse cells respectively. We used lentiviral vectors to co-express a sgRNA with mCherry, as well as a 3XFLAG-tagged Cas9 nuclease 2A-peptide linked to eGFP. Transduced cells were FACS-enriched for eGFP + mCherry + cells after 48 h from which gDNA was extracted from pools after 6 d for a T7 Endonuclease I assay to detect Cas9-directed DNA cleavage. We also isolated single eGFP + mCherry + cells by FACS into 96-well plates and expanded them before isolation of genomic DNA and whole cell lysates. Genomic DNA was isolated using the Purelink Genomic DNA Extraction kit (ThermoFisher) and PCR primers were used to amplify across the targeted region (see Supplementary Table 1). PCR amplicons were denatured and re-annealed to allow heteroduplex formation, then digested with T7 Endonuclease I (NEB) according to manufacturer's instructions and then resolved using DNA gel electrophoresis. We PCR-amplified CTCF exon 3 from genomic DNA isolated from K562 clones, which had been subjected to CRISPR/Cas9-mediated gene editing, using Platinum Taq (Thermofisher). Amplicons were ligated into pGEM-T-Easy (Promega) and then transformed into E. coli. Each clonal amplicon was then confirmed using Sanger sequencing in both directions.

Isolation of mouse embryonic fibroblasts
Ctcf +/mice were obtained on a mixed C57Bl/6:129SvJ background from the Fred Hutchinson Cancer Research Centre (Seattle, Washington) [2]. These mice have had the complete coding region of one Ctcf allele replaced with a loxP-flanked cassette containing a pgk promoter and neo gene, designated Ctcf +/pgkneo ( Figure 3A). Mice homozygous for this allele (Ctcf pgkneo / pgkneo ) exhibit embryonic lethality prior to embryo implantation [2]. Mice were backcrossed at least 10 generations onto C57Bl/6 mice from the Animal Resources Centre (Perth, WA) before beginning phenotyping studies. Timed matings were performed with Ctcf +/male mice and C57Bl/6 females and female mice were checked daily for vaginal plugs. At 13.5 dpc, pregnant females were euthanised by CO2 asphyxiation. The uterine horns were removed and the foetuses released whilst immersed in PBS.
Each pup was removed from its amniotic sac, decapitated and fetal liver removed. The carcasses were minced with a scalpel and then incubated in trypsin/EDTA solution (Invitrogen). The tissue fragments were triturated to break up clumps, and then concentrated using centrifugation to remove trypsin. Fresh trypsin was added to create a homogeneous solution of cellular material. The trypsin was inactivated in excess DMEM medium containing 10% (v/v) FCS and then the centrifugation step repeated. The MEFs were plated in 15 cm plates and allowed to grow for 2-3 d until there were sufficient adherent cells for cryopreservation. The remaining MEFs were transduced with MSCVp53.1224 retroviral supernatant for immortalisation. Cells were selected in 50 μg/mL hygromycin to enrich for immortalised MEFs and then frozen down after 1-2 passages (P2-P3 MEFs). Ctcf +/mice and MEFs were genotyped according to primers listed in Supplementary Table   1.

Western blot analysis
Protein extracts were prepared with cell lysis buffer containing 20 mM Tris-HCl (pH 8), 150 mM NaCl, 1% (v/v) Triton X-100, 0.1% (v/v) SDS, 0.5% (w/v) sodium deoxycholate and EDTA-free protease inhibitor cocktail (cOmplete, Roche Life Science), prior to separation using denaturing sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE). Proteins were transferred onto PVDF membranes in a semi-dry transfer apparatus before immunoblotting. Membranes were blocked in PBS/0.1% (v/v) Tween 20 containing 20% (v/v) BlokHen (AvesLab) or PBST containing 0.3% (w/v) BSA, 1% (w/v) polyvinylpyrrolidone and 1% (v/v) PEG (mw 3350). Protein expression was detected using primary antibodies followed by washing and staining with appropriate secondary antibodies conjugated to horseradish peroxidase (HRP). The HRP substrate SuperSignal Chemiluminescent Substrate (Pierce) was detected on a Kodak Imagestation 4000R Pro or BioRad Chemidoc Touch. Blots were stripped with ReBlot Plus (Merck Millipore) prior to re-probing with protein loading control antibodies. Densitometric analysis of bands from 3 independent blots was performed using ImageJ. Plates with lids removed were placed in a Stratalinker UV Crosslinker (Stratagene) and exposed to UVC irradiation (2,000 μJ for MEFs, 4,000 μJ for K562 cells) and allowed to recover for 18 h. Cells were harvested and stained with anti-Annexin V-APC (BD) according to the manufacturer's protocol and with propidium iodide (PI) solution (5 μg/mL). Cells were analysed on a Fortessa flow cytometer (BD) with analysis performed using FlowJo 9.7.6 software (Treestar).

Cell assays
Cell viability was measured after addition of PI or DAPI (2 μg/mL) and then analysed by flow cytometry. The viable population represents the Annexin V -PIcells; the apoptotic population represents the Annexin V + PIand Annexin V + PI + cells combined.

Bioinformatics analysis
Gene expression and somatic mutation data from the uterine corpus endometrial carcinoma dataset [28] was downloaded from cBioPortal. Of the 500 samples described, 240 contain matched sequencing and copy number alteration data. Statistical tests already conducted on these data were also downloaded including Student's test (p) and Benjamini-Hochberg adjusted p-values (q).
The CTCF-altered gene signature used in subsequent analysis includes all differentially expressed genes (q<0.05).

Results
We used shRNA knockdown to model the cellular consequences of reduced CTCF expression in K562 cells. Western blots showed that CTCF protein expression was significantly knocked down by ~80% in sh.CTCF K562 cells compared to non dox-treated cells and sh.control cells (Figure 1Ai & ii). Cellular proliferation showed that CTCF knockdown resulted in a significant reduction of proliferation (p<0.0001, Figure 1B). We similarly observed a significant reduction in sh.CTCF K562 colony number compared to non dox treatment and to sh.control (both p<0.0001, Figure 1C).
CTCF knockdown led to growth arrest with an increase in G1 phase (p<0.0001), and a concomitant reduction of cells in S (p<0.0001) and G2/M phases (p=0.0036, Figure 1D). We next examined the response of sh.CTCF K562 cells after UV insult and observed CTCF knockdown in K562 cells resulted in an increase in cell viability after recovery from UV exposure (p=0.004) and a decrease in Annexin V-positive cells (p=0.0006, Figure 1E). These data and our previous studies indicated that CTCF dosage is critical for its tumour  Figure 2B. CTCF protein expression was decreased by approximately 50% in most surviving clones irrespective of the sgRNA used ( Figure 2C). As each clone should contain at least one edited CTCF allele, we PCR-amplified the edited region in CTCF, cloned the PCR products and then sequenced them. We detected a mixture of CTCF alleles arising in clones including frameshifts induced by deletion or insertions near the protospacer adjacent motif (PAM) or in-frame deletions leading to microdeletions in the CTCF protein ( Figure 2D). In some clones, we observed three distinct edited CTCF alleles, consistent with K562 cells having a hypotriploid karyotype [36]. SgRNAs #2 and #3 induced 100% and ~96% gene editing efficiency respectively with a ~50:50 mixture of frameshifts and in-frame deletions ( Figure 2D).
We next performed MTT cell proliferation assays on eGFP + mCherry + K562 cell pools (Supplementary Figure 1B) and showed that cells targeted using CTCF sgRNAs #2 and #3 exhibited reduced cellular proliferation (p=0.014 and p=0.012 respectively, Figure 2E). We also performed clonogenicity assays and confirmed that CRISPR/Cas9-directed genome editing of CTCF inhibited colony-forming ability by ~30 % for sgRNAs #2 and #3 ( Figure 2F). Therefore, inducing genetic lesions in CTCF leading to haploinsufficient levels of CTCF in K562 cells had a negative impact on cellular growth.
We then examined Ctcf heterozygous mice to better determine what impact heterozygous deletion of the Ctcf locus has on post-natal growth and development. We backcrossed these Ctcf +/pgkneo mice, which were originally described on a mixed 129SvJ:C57Bl/6J background and exhibited embryonic lethality as homozygotes [2], onto C57Bl/6J mice for at least 10 generations. Backcrossed C57Bl/6J Ctcf +/pgkneo mice bred with wild type C57Bl/6J (WT) mice had smaller mean litter sizes than from normal WT x WT mice ( Figure 3B). This was explained by both female and male Ctcf +/pgkneo mice being born at sub-Mendelian ratios (~28% and ~24% respectively) compared to WT (Ctcf +/+ ) mice (both p<0.0001 Chi-square test; Figure 3C & 3D). After weaning at approximately day 21 we recorded mouse weights 3 times a week until 12 weeks of age. Female Ctcf +/pgkneo mice were smaller than WT littermates up to 7 weeks of age (~14% less body weight, Figure 3E), whereas male Ctcf +/pgkneo mice were consistently smaller than WT littermates in the first 12 weeks of age (~12% less body weight, Figure 3F). This reduced weight phenotype was maintained in male Ctcf +/pgkneo mice even beyond 2 years of age ( Figure 3F). These data show Ctcf haploinsufficiency can significantly impact growth and development in mice. Examination of genetic variation occurring in CTCF in humans using the ExAC database [37] revealed that CTCF is extremely intolerant to genetic variation within the protein-coding region. CTCF exhibits significantly fewer nonsynonymous variants than expected (z score=4.86) and can be classified as haploinsufficient due to intolerance to heterozygous loss-of-function variation (pLI score=1.0; Figure 3G). Two genome-wide CRISPR screens in diploid cells [38], [39] and a synthetic lethal gene trap screen in haploid cells [40] identified 916 core fitness genes essential for cell viability common to all 3 screens, including CTCF ( Figure 3H). These data confirm CTCF as an essential gene in higher order eukaryotes.
To examine the cellular consequences of Ctcf haploinsufficiency, we isolated mouse embryonic fibroblasts (MEFs) from a single litter containing 4 Ctcf +/pgkneo and 3 WT pups. These MEFs were immortalised with a retrovirus encoding stable shRNA knockdown of p53 and then analysed by immunoblot for Ctcf protein expression ( Figure 4Ai). Densitometric analysis of the Ctcf +/pgkneo and WT MEF samples confirmed Ctcf protein was reduced in Ctcf +/pgkneo MEFs to a mean of 58% of WT (p=0.033, Figure 4Aii). We performed MTT assays and showed immortalised Ctcf +/pgkneo MEFs exhibited an increase in cellular proliferation compared to WT MEFs (p=0.0028 day 2, p<0.0001 day 3, Figure 4B). Ctcf +/pgkneo MEFs also displayed an increase in colony-forming ability compared to WT (p<0.0001, Figure 4C). We analysed cell cycle kinetics and showed that Ctcf +/pgkneo MEFs exhibited a decrease in G1 phase compared to WT MEFs (p=0.0072) with a concomitant increase in G2/M phase (p=0.0043) ( Figure 4D). We next examined the cellular response to UV-induced apoptosis and found that immortalised Ctcf +/pgkneo MEFs exhibited a decrease in viability and concomitant increase in Annexin V-positive cells compared to WT cells (p=0.0002 and p=0.0005 respectively) ( Figure 4E). Immortalised p53-deficient Ctcf heterozygous MEFs exhibit pro-tumourigenic characteristics, indicating that Ctcf is acting as a haploinsufficient tumour suppressor gene.
We next used CRISPR/Cas9-mediated targeting of Ctcf in monoallelic Ctcf +/pgkneo MEFs to assess the impact of inducing potentially deleterious Ctcf genetic lesions. Four sgRNAs were designed to target the first coding exon (exon 3) of Ctcf, as well as a control sgRNA targeting the Rosa26 locus (Supplementary Figure 1C). Efficient Cas9-directed cleavage of Ctcf exon 3 was observed using each sgRNA against Ctcf, whereas Rosa26 sgRNA had no detectable effect ( Figure 5A). A clonogenicity assay was performed using Ctcf-targeted eGFP + mCherry + Ctcf +/pgkneo MEFs which showed that colony-forming capacity was significantly reduced to ~30-40% of control ( Figure 5B).
We isolated individual clones for each Ctcf sgRNA by FACS and then examined Ctcf protein expression. Ctcf expression in surviving clones was maintained despite attempts to inactivate the hemizygous Ctcf allele, however, in some clones lower molecular weight Ctcf species were  Figure 5C). These data confirm that CTCF is essential in somatic cells and that CTCF nullizygosity cannot be sustained in viable cells.
To ascertain the impact of CTCF haploinsufficiency in the context of human cancers, we examined a uterine corpus endometrial carcinoma (UCEC) dataset from The Cancer Genome Atlas, which exhibits CTCF genetic alterations in 45 out of 232 patient samples (~19 %) [28]. GISTIC analysis of this cohort assigned each patient sample into potential somatic copy number alterations based on relative CTCF expression level ( Figure 6A). CTCF expression was decreased in a substantial proportion of endometrial cancer samples, some of which can be directly attributed to genetic deletion of CTCF (deep deletion). Many inactivating nonsense and frameshift mutations in CTCF were found in the notionally diploid population (40 out of 179, 22.3%; Figure 6A). Samples with inactivating mutations and confirmed deletions are classified herein as 'CTCF-altered'. We then analysed other gene mutations that co-occurred with or were mutually exclusive in CTCF-altered endometrial cancers. TP53 mutations (66 out of 68) occurred with mutually exclusivity to CTCF mutations (p=9.28x10 -6 , Figure 6B, C); whereas mutations in MED13L, encoding a subunit of the Mediator transcriptional co-activation complex, co-occurred with CTCF mutations in 13 out of 23 cancers (p=2.64x10 -5 , Figure 6B).
We analysed RNAseq data available for these endometrial cancer samples and showed that CTCF gene expression was not significantly decreased in CTCF-altered cancers despite the presence of inactivating mutations (Supplementary Figure 3A). We next examined the chromosomal distribution of all expressed genes detected above threshold in endometrial cancers (~13,000) and found an enrichment for genes expressed on chromosomes 11, 16, 22 and X ( Figure 6D). However, in CTCF-altered cancers there was enrichment for genes expressed on chromosomes 1, 7, 9, 17 and 20 ( Figure 6C)

Discussion
Haploinsufficiency arises when only a single functional copy of a gene is inadequate for normal cell orientation of CTCF target sequences to alter genome topology [17]. Unbiased genetic screens using CRISPR/Cas9 sgRNA libraries or synthetic lethality in haploid cells have identified CTCF as part of an 'essentialome' containing ~900 core fitness genes required for cell viability [38]- [40].
However, only two studies to date have directly focused on targeting CTCF in mammalian cells using CRISPR/Cas9 editing [48], [49], but with disparate outcomes. CTCF heterozygous MCF10A clones generated by CRISPR exhibited similar proliferation rates compared to control, though cells were slower to repair double-stranded DNA breaks [48]. Silencing of CTCF in the RS4;11 acute lymphoblastic leukaemia cell line increased colony numbers in soft agar overlays compared to control, though CTCF protein appeared to be reduced to minimal levels [49].
Our strategy was to examine the essentiality of CTCF in somatic cells by CRISPR/Cas9 targeting of hemizygous MEFs that only contain one coding allele of Ctcf, we were unable to completely abolish Ctcf protein expression using CRISPR. Accordingly, in surviving clones we detected a high incidence of in-frame microdeletions in Ctcf. As these microdeletions occur in the intrinsically disordered N-terminus we do not expect them to significantly impact Ctcf function. Induced frameshift deletions were also likely to produce truncated CTCF proteins initiating from alternate in-frame ATG start codons within the N-terminus. These results confirm that CTCF is absolutely required for somatic cell viability and that CTCF cannot be completely inactivated in cells.
Interestingly, residual CTCF protein levels can be depleted to minimal amounts in the cell before viability is significantly impacted. Targeted degradation of CTCF in mouse embryonic stem (ES) cells using an auxin-inducible degron system highlighted that endogenous CTCF protein levels could be decreased by up to 99% for at least 2 d duration without a significant impact on cell proliferation or viability [50]. Acute depletion impacted on CTCF looping and insulation of TAD regions, but genomic compartmentalisation was maintained [50]. CTCF-null ES cells can progress to the blastocyst stage (E3.5) purely via retention of maternal CTCF mRNA, but exhibit periimplantation lethality by E4.5-E5.5 [2].
We have also shown that CTCF has a dose-dependent impact on embryonic development, as even haploinsufficient levels of Ctcf protein affected embryonic development in mice. We observed heterozygous Ctcf mice being born at sub-Mendelian rates (24-28%) compared to WT littermates, which was previously suggested by a study using mice with a conditionally targeted Ctcf allele, but was not thoroughly quantified or statistically verified [3]. Interestingly, the same Ctcf hemizygous mice are born at normal Mendelian rates on a mixed C57Bl/6J:129SvJ background [6], indicating that the strain background is important factor to consider in any Ctcf genetic deficiency studies in mice. Ctcf heterozygosity also impaired normal mouse weight gain during adult development for up to 7 weeks, and which then remained constant in aged male mice. As these mice were fed a normal chow diet, we could not determine whether Ctcf heterozygosity impairs body weight control, metabolism or nutrient signaling pathways. In future studies we will examine glucose and insulin levels in plasma after feeding-fasting cycles, body tissue composition using dual energy x-ray absorption as well as studying the impact of different chow compositions on Ctcf heterozygous mouse development.
CTCF hemizygous mice are more susceptible than WT mice to spontaneous cancer development, as well as radiation-and chemically-induced cancers [6]. Tumours in Ctcf +/mice compared to WT mice also exhibit increased aggressiveness in terms of invasion, metastatic dissemination and mixed epithelial/mesenchymal differentiation, confirming CTCF as a haploinsufficient tumour suppressor [6]. Our current findings showing an increase in cell proliferation, colony forming ability and numbers of cycling cells in p53-shRNA immortalised Ctcf +/-MEFs support these conclusions. window surrounding divergent CpGs exhibiting a generalised pattern of DNA hypermethylation [6].
In humans with heterozygous CTCF mutations exhibiting an intellectual disability, specific CTCF sites exhibited DNA hypermethylation [47]. This epigenetic dysregulation may offer an explanation as to why differential gene expression was observed at particular chromosomal loci in CTCF-altered endometrial cancers.
One possible hallmark of CTCF-altered endometrial cancers is the downregulation of tumour suppressor genes including PIK3CA, CDKN2A, CDH6 and IGF2BP2. The tumour suppressor PIK3CA is ranked fifth after CTCF in the most frequently mutated genes in UCEC [6] whilst CDKN2A is downregulated in POLE, MSI, and copy-number low cancers compared to high-copy number cancers [28]. CDH6, which helps maintain epithelial integrity in the endometrium [52], has been shown to be a putative tumour suppressor in cholangiocarcinoma [53]. IGF2BP2, which was the most down-regulated gene in CTCF-altered endometrial cancer, was identified as a candidate tumour suppressor gene in a pan-cancer screen for homozygously deleted genes [54]. Loss of IGF2BP2 staining, which is a feature of endometrioid cancers but not serous cancers, has been proposed as a biomarker for distinguishing endometrial tumour pathology [55].
A second hallmark of CTCF-altered endometrial cancers is the upregulation of estrogen-responsive genes, which includes KIAA1324, MLPH, MSX2, SPDEF, TFF3 and PIGR. CTCF mutations do not occur in a tumour type-specific manner, but rather they define a subset of hormone-responsive cancers [31]. CTCF is a negative regulator of the pioneer factor FOXA1, which facilitates estrogen receptor interactions with chromatin in response to estrogen [56], [57]. Therefore in CTCF haploinsufficient endometrial tumours, FOXA1/ER interactions with chromatin may increase leading to upregulation of estrogen-responsive genes. KIAA1324, which is a positive regulator of the autophagy pathway and may protect cells from cell death, was the most upregulated gene in CTCF-altered endometrial cancer [58]. KIAA1324 is a marker of grade I endometrial cancer which decreases with increase in tumour grade and disease stage [59] and is a key member of gene signatures classifying histological subtypes [41], [42]. Other estrogen-responsive genes upregulated in CTCF-altered cancers, namely SPDEF, TFF3 and PIGR, are also components of these gene signatures, indicating that loss of CTCF could be an important factor determining endometrial cancer progression and pathology.

Conclusions
We examined CTCF essentiality and haploinsufficiency in somatic cells and mice using various molecular genetic techniques and models. Despite achieving efficient genome editing of CTCF using CRISPR the inability to obtain complete ablation of CTCF expression reinforces its requirement. In all cases, cellular fitness in CTCF-targeted cells was comprised leading to surviving cells compensating with reduced CTCF protein expression or truncated CTCF protein variants.
Consequently cell proliferation, colony-forming ability and cycling cells were reduced. However, in the presence of additional genetic hits, such as in p53, CTCF haploinsufficient cells exhibited known cancer hallmarks, namely increased proliferation and reduced cell cycle control. In human endometrial cancer datasets, we identified a unique gene signature in CTCF haploinsufficient cancers arising from differential gene expression at specific loci. Downregulation of tumour suppressor genes and upregulation of estrogen-responsive genes may be a molecular feature of CTCF-altered endometrial cancers. Our study clearly demonstrates that CTCF is a haploinsufficient tumour suppressor gene that is essential for somatic cell viability and protects against cancer. As the master of weaver of the genome CTCF plays an essential role in chromatin organisation, the full impact of CTCF haploinsufficiency on three-dimensional chromatin architecture remains to be elucidated.

Ethics Approval
All animal experiments were performed in accordance with an approved institutional animal ethics protocol from the Royal Prince Alfred Hospital Animal Welfare Committee (SSWAHS #2013/046).

Availability of data and material
All data generated or analysed during this study are included in this published article (and its supplementary information files).         (11) (11) (6) (11) (17)